Fix signature of torch allocator callbacks #1408
Closed
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
The deallocation function now also takes the device id.
Since both halves of the pair now receive the device on which to perform the (de)allocation, we switch from using
get_current_device_resource to using the (more correct) get_per_device_resource. This necessitates a workaround in Cython: rmm::cuda_device_id has no nullary constructor, and so cannot be stack-allocated the way Cython transpiles code. Instead perform a heap allocation and then delete it.
Checklist